We have established a large dataset of over 450,000 application packages from various colleges and universities, which includes applicants’ essays, socio-demographics, standardized test scores, and more. To assess the creative potentials of applicants, we developed computational creativity metrics using semantic distance methods and large language models (LLMs) to analyze creativity in college admissions essays. We then tested whether our creativity metrics could predict future success in various areas, including GPA, graduation rates, and career outcomes. Additionally, we aim to explore how incorporating creativity as a criterion in high-stakes admissions decisions can help reduce disparities in higher education.

Our computational metric showed a strong correlation with human experts' creativity ratings of admissions essays. Applicants who wrote more creative essays, as evaluated by our metric, achieved higher GPAs in college and had lower rates of D, F, or withdrawal, even after accounting for standardized test scores. Notably, our creativity metric was much less associated with sociodemographic factors, such as race and ethnicity, compared to standardized test scores. We replicated our findings using data from four universities with varying characteristics, ensuring the robustness of our results. We plan to expand our research to explore the value of creativity in the context of the increasing use of generative AI.